Ph. D. Project
definition of 3D digital twin for logistics containers : an optimization problem
2020/12/01 - 2023/11/30
Other supervisor(s):
ADDIS Bernardetta (
The Industry of Future Concept aims to provide a framework for business development. Many regional or national initiatives have already been initiated to support the economic development of the actors concerned towards this new concept. The work proposed in this thesis is part of this same approach.
The industry of the future is based on key elements [Panetto, 2019] that can foster the development of tangible goods companies. Among these elements, the concept of Digital Twin (or Digital Twin) is intended to represent virtually a physical object, a process or a system and its properties. This digital replica, with potentially multiple uses, is continuously fed with data collected on the real physical object for better control of it.
This thesis focuses on the concept of Digital Twin in the particular context of current and future logistics. More specifically, the objects of interest in this context are, because of their important dynamics, the containers of packages that are grouped, sorted, unbundled and shipped. The twin three-dimensional model must also evolve dynamically over time.
The construction of a three-dimensional model of container and its contents (packages) shares some properties with packing problems [Addis, 2008], but in the case of our application target, the problem is "reversed" : the packing is given (even if it is unknown) and some information on its configuration is available (for example the size of the packages, the neighborhood between parcels, etc.); the objective is to reconstruct, using this partial information, the corresponding arrangement of the packages (packing).
This description highlights an important element, namely the need to gather information about the system in an indirect way. The quality and quantity of this information influences the difficulty (possibility) of the reconstruction problem. The key point of this component is to design a collection system that produces "good" information, to allow its integration into the physical system. Previous work exists on the subject (see for example [Thiesse, 2006], [Zhou, 2015], [Abbate, 2009]), but the instrumentation they propose is not economically viable for many logistical applications.
Hoa Tran Dang's thesis work [Tran Dang, 2017a] has shown that, under certain hypotheses, it was possible (via a given instrumentation of the WSN type) to obtain a 3D model of the contents of a container. The problem is modeled as a packing problem with additional constraints from the WSN neighborhood graph. It is solved by an approach inspired by constraint programming [Focacci, 2004] [CP, 2006]. However, the assumptions made in this work do not always make it possible to obtain a single arrangement solution corresponding to the actual arrangement (several 3D configurations may give rise to the same neighborhood graph). Indeed, the information collected by the sensor network considered is not sufficient to give rise to a single digital twin even under the reductive assumptions taken into account (chirality, package in the form of a parallelepiped, accelerometer in each node, etc. .).
Reducing the number of solutions proposed from the measured data is the main barrier to overcome for obtaining a unique digital twin. The real industrial application context makes it necessary to take into account economic and technological constraints in order to arrive at viable and relevant solutions. All the special features of this problem make it unique and new. To remove this lock, the following research tracks will be investigated:
1. Formalize and integrate in the pre-existing model [TranDang, 2017b] static mechanical constraints (balance of packages) to complete the information of the sensors by knowledge not yet exploited; extend / modify the resolution method to incorporate the ability to handle strongly nonlinear constraints.
2. Qualify, quantify and place an additional instrumentation to that already deployed, and at minimum cost and / or energy to obtain additional data (optimization models and resolution algorithms)
3. Study techniques that enable the digital twin to gather information complementary to that available by the sensor network, autonomously, asynchronously and intelligently, to ensure its integrity at all times, while respecting the constraints of cost and efficiency. energy consumption of the system.

Parallel to these first tracks, and in a completely reversed approach, we propose to study a second approach. The digital twin is the starting point: it is considered unique. The first question we have to answer is: Is it possible to represent, in a formalism to be determined (e.g. [Buchsbaum, 2008]), and without any ambiguity, this arrangement? If the answer to the previous question is positive, it will then be necessary to determine if there is at least one way to instrument the physical system to instantiate the proposed formalism to the different possible 3D arrangements.
If successful on this second approach, it should be compared with the original approach, based on the respective costs of each.
From a scientific point of view, the project is based on different disciplinary fields covered by the reciprocal skills of the co-supervisors:
• Instrumentation technologies for objects of interest (parcels / containers) and data communication for the collection of these objects.
• Numerical optimization to allow the modeling of the problems of the mentioned decisions and to propose efficient methods of resolution.
The scientific partnership in the framework of this thesis between the CRAN (instrumentation, logistics) and the LORIA (optimization) is linked to the strong interactions and interweaving between each of the lines of research to be conducted in order to propose an innovative and applicable solution in the field industrial.
digital optimization, digital twin , containers, instrumentation, data collection
Bonnes connaissances en RO et en optimisation nécessaire
Eco-Technic systems engineering